What's New
Relevance-Aware Synthesis
score_result_file()— 4-factor composite scoring (word count, code blocks, specificity, structure) for ranking provider resultsrank_results_by_signals()— Quality-based ranking before synthesis (rank, don't delete)- Enhanced
aggregate_results()— Annotated quality scores, structured synthesis output (Key Findings, Detailed Analysis, Conflicts & Trade-offs, Recommendations), minority opinion preservation - Enhanced
synthesize_probe_results()— Re-ranks results, 6-section structured output with Priority Matrix
CC Pre-Prompt Alignment (v2.1.72)
- 8 new detection flags: ExitWorktree, Agent model override, effort redesign, DISABLE_CRON env, hidden HTML comments, bash allowlist v2, /clear preserves bg, team model inherit fix
- Wired: cron disable during embrace/parallel, effort symbols ○◐●, model_override_supported in Agent Teams JSON
- Total flags: 80 across 27 version_compare blocks
Model Catalog & Resolution
- Centralized model catalog with provider→model mappings and capability metadata
resolve_octopus_model()v3.0 — Env var → config file → capability → phase routing → tier mapping → session override → catalog default- In-memory + persistent file caching for resolved models
- Health check integration:
check_codex_auth_freshness()in preflight
Post-Run Usage Reporting
- Token usage and cost estimation displayed after multi-provider workflows
- Per-provider breakdown with model-specific pricing
Test Fixes
- Fixed 6 pre-existing test failures across security, model resolution, opus integration, and perplexity tests
- 70/70 tests passing (pre-push hook verified)
Full Changelog: v8.48.0...v8.49.0